BIANCA‐MS: An optimized tool for automated multiple sclerosis lesion segmentation

Abstract

In this work we present BIANCA‐MS, a novel tool for brain white matter lesion segmentation in multiple sclerosis (MS), able to generalize across both the wide spectrum of MRI acquisition protocols and the heterogeneity of manually labeled data. BIANCA‐MS is based on the original version of BIANCA and implements two innovative elements: a harmonized setting, tested under different MRI protocols, which avoids the need to further tune algorithm parameters to each dataset; and a cleaning step developed to improve consistency in automated and manual segmentations, thus reducing unwanted variability in output segmentations and validation data. BIANCA‐MS was tested on three datasets, acquired with different MRI protocols. First, we compared BIANCA‐MS to other widely used tools. Second, we tested how BIANCA‐MS performs in separate datasets. Finally, we evaluated BIANCA‐MS performance on a pooled dataset where all MRI data were merged. We calculated the overlap using the DICE spatial similarity index (SI) as well as the number of false positive/negative clusters (nFPC/nFNC) in comparison to the manual masks processed with the cleaning step. BIANCA‐MS clearly outperformed other available tools in both high‐ and low‐resolution images and provided comparable performance across different scanning protocols, sets of modalities and image resolutions. BIANCA‐MS performance on the pooled dataset (SI: 0.72 ± 0.25, nFPC: 13 ± 11, nFNC: 4 ± 8) were comparable to those achieved on each individual dataset (median across datasets SI: 0.72 ± 0.28, nFPC: 14 ± 11, nFNC: 4 ± 8). Our findings suggest that BIANCA‐MS is a robust and accurate approach for automated MS lesion segmentation.

LST-AI: a Deep Learning Ensemble for Accurate MS Lesion Segmentation

“…With regard to clinical applicability of automated lesion segmentation tools, the sensitivity is crucial as diagnosing and monitoring MS relies on the detection of (new) lesions. A newly published method, namely BIANCA-MS (Gentile et al, 2023), has also been validated using the MICCAI 2016 test dataset and yielded results similar to ours in terms of DSC and false positives (in terms of lesion detection). However, the median number of false negatives was equal to 11(IQR: 18) for BIANCA-MS, whereas LST-AI yields a median number of false negatives equal to 4 (IQR: 8), again highlighting the high sensitivity of our proposed method towards lesion detection.…”

“…However, assessing generalizability of segmentation models requires validation on external datasets. This has been done in recent studies, which used different train and test set pairings, including in-house and publicly available data such as ISBI 2015 and MICCAI 2016 data (e.g., train on in-house data and test on MICCAI 2016 data) (Billot et al, 2021; Cerri et al, 2021; Gentile et al, 2023; Kamraoui et al, 2022; X. Li et al, 2022; McKinley et al, 2021; Rakić et al, 2021).…”

“…An important CNN-based architecture is the UNet, which has been applied in many previous lesion segmentation studies (Ashtari et al, 2022; Hashemi et al, 2022; Krishnan et al, 2023; La Rosa et al, 2020; Ronneberger et al, 2015). Furthermore, recent studies successfully train their models on one dataset and test it on another, external dataset, for which the MICCAI 2016 (Commowick et al, 2021) and ISBI 2015 (Carass et al, 2017) datasets are often selected (Cerri et al, 2021; Gentile et al, 2023; Kamraoui et al, 2022; Krishnan et al, 2023; X. Li et al, 2022; McKinley et al, 2021).…”

“…1101/2023.11.23.23298966 doi: medRxiv preprint approaches that have previously been published are optimized towards transferability: Valverde et al have provided nicMSlesions, a CNN-based lesion segmentation method that is able to adjust to a new image domain by retraining their model on a single image (Valverde et al, 2019). Furthermore, recent studies successfully train their models on one dataset and test it on another, external dataset, for which the MICCAI 2016 (Commowick et al, 2021) and ISBI 2015 (Carass et al, 2017) datasets are often selected (Cerri et al, 2021;Gentile et al, 2023;Kamraoui et al, 2022;Krishnan et al, 2023;X. Li et al, 2022;McKinley et al, 2021).…”